Berlin 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 62: Developement of Calculation Methods III
MM 62.3: Talk
Thursday, March 21, 2024, 16:15–16:30, C 264
Machine Learning Potentials for Multi-State Systems: Predicting Photoluminescence Spectra from Molecular Dynamics — Christopher Linderälv1, 2, •Nicklas Österbacka2, Julia Wiktor2, and Paul Erhart2 — 1University of Oslo — 2Chalmers University of Technology
Divacancy defects in 4H-SiC show potential as single-photon emitters, which are important devices in quantum information technology. The photoluminescence spectra of these defects are crucial for this application, and accurate prediction of such spectra can aid in both understanding of the underlying defect physics and in device optimization.
We introduce a method for predicting photoluminescence spectra from the energy difference between the ground and excited state of defects sampled from molecular dynamics (MD) simulations, for which we employ machine learning potentials (MLPs) to extend the accessible length and time scales. Standard MLP construction approaches lead to exponential divergence in absorption and emission energies with increasing system size, however. To circumvent this, we introduce a method for the construction of MLPs capable of simultaneously describing both states.
We construct such a potential for a 4H-SiC divacancy defect and show that our MD-based approach yields emission spectra in good agreement with the generating function approach, which is the gold standard for such predictions from atomic structure. We also highlight the advantages of our MD-based method, emphasising synergies with MLPs.
Keywords: defects; quantum; machine learning; molecular dynamics; photoluminescence